Improved Absolute Quantification using Bayesian Penalized Likelihood Reconstruction on a Digital PET/CT – Towards True Uptake Measurement

To evaluate the quantification accuracy, we measured the recovery coefficient (RC) and contrast 3 recovery (CR) in a phantom study and used it as a guide in clinical evaluation about digital positron 4 emission tomography -computed tomography (PET/CT). The RC and CR of the PET images reconstructed with 4 different methods [ordered subsets 7 expectation maximization (OSEM), time of flight (TOF), TOF-point spread function (PSF), bayesian 8 penalized likelihood (BPL)] were measured in the phantom study. And, SUVmax and SUVmean (the 9 maximum and mean of the standardized uptake values, SUVs) of 75 small pulmonary nodules 10 (sub-centimeter group: <10 mm and medium size group: 10 - 25 mm) from 26 patients were measured 11 with those methods. For lesions smaller than 3 times of the spatial resolution, partial-volume-effect 12 correction (PVC) based on RC values was performed.

3 pathological processes in vivo. In the development of PET imaging system, many research efforts have 31 been devoted to reinforce the power of quantification accuracy and small lesion detectability [5,6]. On 32 one hand, due to the partial volume effect caused by the limited spatial resolution of conventional PET 33 systems, the radiotracer uptake is usually underestimated when measuring a lesion smaller than the 3 34 times of the spatial resolution [7]. On the other hand, the traditional iterative image reconstruction 35 algorithm, e.g. ordered-subsets expectation maximization (OSEM), is not able to reach full 36 convergence because the noise in the image grows with each iteration, and hence there exists a 37 compromise between iteration and noise resulting in partial convergence [8,9]. Therefore, it is 38 particularly difficult to assess the true metabolic activity of the small lesions, such as sub-centimeter 39 pulmonary nodules or lymph nodes. Many studies demonstrated the challenges of evaluating 40 sub-centimeter nodules, for instance, the differential diagnostic sensitivity became lower for 41 determining malignancy of such nodules than that of larger ones [10][11][12], and even false-negative 42 findings can be generated [13]. The visibility of small lesions and quantification accuracy facilitates the 43 improved staging, treatment planning, response monitoring and prognostic estimation, and they are of 44 importance for clinical diagnostic confidence and patient management [14].

45
Generally, it is impossible to know the true uptake of a lesion in vivo. In order to access the 46 quantification accuracy, the most reliable method is to measure the recovery coefficient (RC) in the phantom study, which gives the true activity by calculating the ratio of measured activity to the true one 48 determined by the dose calibrator [15]. Contrast recovery (CR) is another important index reflecting the 49 true uptake ratio in the lesion and background by showing the ratio between measured 50 sphere-to-background activity and known sphere-to-background activity [16]. The BPL based 51 reconstruction algorithm (known as Q.Clear, GE Healthcare) on a digital PET/CT system (Discovery 52 MI, GE Healthcare) has shown significant advantages over conventional reconstruction algorithm 53 (OSEM) in photomultiplier tube (PMT)-based PET/CT scanners [17][18][19][20]. It has been revealed that the 54 new BPL-based reconstructed PET images provided significant increases in signal-to-background, 55 signal-to-noise (SNR) ratios and SUVs, with greatly enhanced visual sensitivity for assessing small 56 pulmonary nodules [21,22], liver metastasis [23] and mediastinal nodes in non-small cell lung cancer 57 [24]. Although SUVs measured from the BPL reconstruction algorithm achieved higher values, 58 whether this SUV elevation reflects higher quantification accuracy need to be verified.

59
Therefore, we evaluated the quantification accuracy of BPL-based reconstruction on SiPM-based 60 4 PET/CT system with a phantom study and performed a clinical study on pulmonary nodules based on 61 the results of the phantom study.

63
Phantom 64 A National Electrical Manufacturers Association (NEMA) image quality phantom [25] was 65 utilized in this study with 6 spheres filled with 13.2 kBq/mL 18 F-NaF in a 4-to-1 ratio (sphere to 66 background activity concentration). Recovery coefficient (RC) and contrast recovery (CR) was 67 measured using the following equations (Eq.1-2).

68
Where AM is the measured activity (in kBq/mL) in each sphere delineated on CT images; AB is the 72 measured activity in the background. AK is the known activity (in kBq/mL) in the sphere; C is the 73 known ratio of activity in the sphere to the background (that is 4:1 in the study).   90 where x is the image estimate; i is the pixel index; yi represents the measured PET coincidence 91 data; P is the system geometry matrix; β is the penalization factor; R(x) is the penalty to control noise.

92
Clinical evaluation

93
For the retrospective nature of this study, informed consent is waived. We retrospectively 94 reviewed the imaging data of all patients with lung nodules who underwent 18        SUVs determined from BPL method were significantly higher than those from other reconstruction 149 methods (P < 0.001 for SUVmax and SUVmean) (Figure 3).

150
151 Table 3 The impact of different reconstruction methods on SUVs of small pulmonary nodules before/after PVC 152 However, the significance was highly affected by the nodules' sizes, as shown in Table 3

167
SUVmean and the corr SUVmean in each reconstruction algorithm (P < 0.001 for all), but also between 168 BPL method and the other three ones in corr SUVmean (P < 0.001 for all). However, differences in 169 SUVs measured from 4 reconstruction methods were not significant for the medium size group (Table 3, 170 P = 0.771 for SUVmax and P = 0.711 for SUVmean).

172
A reliable and precise measurement of radiopharmaceutical uptake is more and more important in

187
For the quantification of pulmonary nodules, significant increases in the SUVmax and SUVmean

188
were observed when reconstructed with BPL-based algorithm compared to the other 3 methods ( Figure   189 3), and the increases for those of sub-centimeter nodules were significantly greater than those of the 190 medium size group (Table 3, P < 0.001) By comparing the changes of SUVs among different 191 reconstruction methods in the medium size group, increments can be found but was not statistically 192 significant (Table 3, P > 0.05). This was because, firstly, medium size group was closer to full approach is on the assumptions that the lesion has a regular spherical shape and uniform distribution of 208 radioactivity [27]. Due to the limited size range of phantom sphere being from 10 mm to 37 mm, the 209 PVE correction for those nodules either smaller than 10 mm or larger than 37 mm should be careful. In

223
The respiratory motion during PET imaging may compromise the spatial resolution, resulting in 224 smearing or blurring effect of PET images, which would eventually affect SUV measurement. Existing 13 researches suggested that respiratory movement leads to overestimation of radiotracer-avid target 226 volume and reduction in the SUV due to the recorded number of coincidence events tends to distribute 227 in a larger volume [30]. The respiratory gating system can be employed in the future study to facilitate 228 the improvement on spatial resolution and quantification accuracy, especially when radiotherapy 229 treatment planning is desired [31][32][33].

230
Furthermore, the small pulmonary nodules in our study were not histopathological verified. The